AI Accelerator for Edge Computing

The U.S. Army's modernization efforts, as outlined in the Army 2030 Visual report, highlight the need for advanced technologies like GSI Technology's Gemini-II® AI Accelerator with Edge Models. The Gemini-II, paired with efficient and precise Edge AI models, addresses key requirements such as low-signature combat forces, protection against various attacks, and enhanced sensing capabilities. Current edge computing technologies face significant bottlenecks and performance limitations, particularly in high-performance computing applications. Traditional architectures like CPUs, GPUs, and FPGAs struggle with I/O bottlenecks, high power consumption, and scalability issues, making them inadequate for the increasing demands of AI applications such as autonomous vehicles, cybersecurity, and data analytics. GSI Technology's Gemini-II introduces a compute-in-memory (CiM) architecture that processes data directly within the memory, reducing latency, power consumption, and increasing processing capacity. This architecture supports parallel processing with two million-bit processors per chip, making it ideal for large-scale sensor data processing and scenarios with limited power availability. The Gemini-II AI accelerator is designed for real-time data processing with lower latency and higher efficiency, suitable for applications ranging from autonomous vehicle navigation to data center computation. By overcoming the limitations of traditional CPUs and GPUs, Gemini-II provides a scalable and efficient solution for AI workloads, making it a key technology for the Army's edge computing needs. In this SBIR Phase I effort, GSI will conduct a feasibility study to integrate the Gemini-II with Edge AI models to solve an Army-specific edge compute use case, ensuring the technology meets the Army's operational requirements.